Feature Requests

Operational Agents will be governed digital workers that monitor, analyze and carry out work across the solutions and tables you grant them. Unlike Field and Workflow Agents, they aren't tied to a single trigger: they watch continuously and complete work end to end, such as escalating overdue findings or flagging SLAs before they breach. Every action runs within least-privilege permissions, with human review and approvals where you need them and a full audit trail.

AI Skills will let you package instructions and optional reference files into a reusable, governed action. Define a Skill once, choose which members can use it, and call it from any agent or Co-Pilot prompt by typing /. Skills run within the user's permissions, so teams can reuse proven work, such as generating a vendor risk assessment, summarizing SOC 2 findings or mapping controls to ISO 27001, consistently across every agent.

SmartSuite Copilot is an AI assistant built into SmartSuite. Describe what you need in plain language and the Copilot builds and changes workspace structure, acts on records, and answers questions about your workspace data, working from what your workspace means, not only from how it is put together.

Today the hosted MCP server exposes one flat list of builder tools: create and change tables, fields, views, automations and records. That is right for a builder, but an end user who connects Claude, ChatGPT or Copilot to SmartSuite wants to run the work, not build it. Request: a second, end-user layer of MCP tools that follows the business logic of each solution, the way the UI does. A toolbox per solution: a small set of named, described tools backed by what already exists in the solution (submit a form, read a saved view, read a dashboard's values, run an automation with typed inputs, create or update records in an allowed field list). Published by the solution manager; nothing exposed by default. Exposed only by the signed-in user's permissions, re-checked on every call, with the same attribution and audit as today. Found through one connector with a small meta layer (list toolboxes, search tools, run a tool), so the assistant loads only what it needs and the builder tools can sit behind a separate Build mode. Acceptance: the same request through the connector returns the same numbers a dashboard shows that user, and a form tool creates the same record with the same validation as the form in the app. Why: assistants connected to a flat builder tool list spend their context on structure and get the business process wrong. Toolboxes let a team expose "approve a purchase request" or "log a risk" without exposing "delete a table".

It would be great to use Openrouter directly to then choose whichever model the team wants or needs for AI. Currently being limited to only the big names and not having access to cheaper inference is enough of a blocker to not use the features and not get the value, especially for teams that are value conscious.

AI Guidance will let you define reusable policy that SmartSuite AI applies automatically, such as rating scales, terminology, disclaimers and escalation rules. Each piece of Guidance can be scoped to specific solutions and applied to Field Agents, Workflow Agents or every AI run, so output stays consistent and on-policy without anyone having to remember to include it.

Enable AI fields to be updated at a preset schedule so it saves excessive updates and unecessary token consumptions. For an AI field that references many fields, instead of updating 4 times a day I can have it update twice a week and save credits. These should be set up within the automation portal.

I have a detailed report on this. Please note the attachment.

Please add explicit trigger controls for AI Field Agents so users can choose exactly which field changes cause an automatically running Field Agent to execute. I originally understood the Run Automatically setting to mean the agent runs whenever any part of the record is updated. SmartSuite's current documentation clarifies that automatic runs actually occur when fields referenced in the AI Field Agent prompt change . That is useful behavior, but it creates a different limitation: fields used as AI context and fields used as execution triggers are currently the same thing. For example, an AI Field Agent might reference both: * Transcription — the source content that should cause the summary to regenerate * Helpful Instructions for the AI — additional context the AI should use Both fields need to be available to the prompt, but a user may want only Transcription to trigger an automatic run. There is currently no explicit setting to distinguish “use this field as context” from “run the agent when this field changes.” Ideally, AI Field Agent settings would include a Run when these fields change option that allows one or more trigger fields to be selected independently of the fields referenced in the prompt. It would also be useful to support trigger conditions such as: * Changes to any new value * Changes to a specific value * Becomes populated * Is cleared This would make AI Field Agent behavior more explicit and predictable, avoid unnecessary AI runs, and support more precise automated workflows. There is also a UX issue related to this. The current tooltip for Run Automatically says: “Allow AI to run automatically when a record is updated” That wording implies a record-level trigger, while the documentation says automatic runs are based on changes to referenced fields. At minimum, the tooltip should be updated to accurately describe the current behavior, for example: “Allow AI to run automatically when a field referenced in the prompt is updated.” Explicit trigger-field configuration would be preferable because it would allow prompt context and execution logic to be controlled independently.

The AI Field Agent feature is fantastic, but for now, we can only add an OpenAI key. It would be very helpful if we could also add API keys for other AI platforms (like Gemini, Azure and Anthropic). This is important for companies that already have a paid plan with a provider, as they need to know that their data will not be used for AI training or shared.